Acute kidney injury (AKI) is a common complication of cardiac surgery with potentially serious effects on long- term health, including need for dialysis and increased risk of cardiovascular events and mortality. Current gaps exist in the diagnosis, prevention and treatment of AKI, and there is great interest in developing AKI risk prediction models to address these issues. Traditional clinical and demographic variables have limited predictive capacity for AKI. As a result, there is increasing interest in using biomarkers and biomarker combinations to predict AKI. While biomarkers have the potential to identify patients at high risk for AKI, issues remain. The proposed research will address two such issues: the use of multi-level AKI outcomes and the development of biomarker combinations in multi-center studies. First, while multi-level AKI outcomes (such as no, mild and severe) are defined, severe AKI is often the outcome of interest, as this outcome is most strongly associated with morbidity and mortality. Multinomial modeling methods and specialized model selection techniques will be used to leverage variation in biomarker levels between individuals with no AKI and those with mild AKI to improve prediction of severe AKI risk. Second, though multi-center biomarker studies typically offer greater power and increased generalizability of results, i is possible to have center differences due to varying AKI prevalence and/or differences in biomarker measurements. Such differences, if ignored, can lead to bias in the assessment of the performance of biomarker combinations. Thus, the proposed research will create methods for developing biomarker combinations in multi-center studies, including tools to characterize differences by center and techniques to identify predictive biomarker combinations that account for center. Addressing these two issues will advance the potential of biomarkers for AKI risk prediction. Biomarkers capable of predicting risk of AKI in the setting of cardiac surgery could be used to reduce the burden of AKI by (1) providing a more accurate diagnosis; (2) diagnosing AKI earlier, opening a therapeutic window; (3) identifying high risk individuals for whom preventative measures should be implemented; (4) enriching clinical trial enrollment, aiding in the development of novel therapies and tools for prevention; and (5) providing the clinician and patient with better information on which to base decisions. Any of these outcomes would transform clinical care in nephrology and improve the health of patients undergoing cardiac surgery.